GIMP: most if not all colour/value transfer functionality is broken
 

I'm a photographer making the switch from Mac to Linux, and currently trying to make GIMP do the work. Alas, there's a serious problem that runs deep: the colour/value processing functionality does not work as it should. You can prove this for yourself with 3 easy experiments:

1.

Open any photo of a decent size and duplicate it into a new layer. Zoom in to 300% or 400% to better see the effects. Desaturate this new layer using the Hue-Saturation command. Set the layer's blending, transfer mode to Value. You won't fail to notice that the desaturation has seriously affected the Value, something that should not happen. The value should have remained the same.

2.

Revert the image to its original and again duplicate it into a new layer. Now we'll approach the problem from the other end and pretend we want to reduce colour noise by blurring the colours only, leaving luma Values the same. So apply a strong blur to the new layer and change the blending mode to Color. Again, the luminance is seriously affected, giving the image an appearance of a greater degree of mottled grain. This should not have happened, the luminance should have remained the same, only the colours should have become smoother and perhaps more muted.

3.

To satisfy ourselves that this is not merely a problem with layers, we tackle the problem in yet another way. Revert the image again, but this time don't duplicate into a new layer, apply a strong blur to the image directly. Now pull down the Fade command from the Edit menu and set the transfer mode to Color, leaving the percentage at 100%. It's no better, the luminance is as badly affected as it was with the layer method.

This leads me to my question: is there an easy way to report this as a bug? I've looked at the Gimp Bugzilla page and can't make sense of it. Would someone experienced here care to repeat the tests and do the report? We could treat it as sharing the workload... ;)

I will look into this as soon as I have time.

Meanwhile, is it possible that GIMP just does not work they way you are used to? (eg if you are accustomed to Photoshop)

Thanks!

I've thought of this, but I don't think it applies. Processing algorithms are at stake, not "the interface I'm used to". If I'm wrong, it should be quite easy to refute what I'm saying.

I'm having trouble coming up with anything conclusive on this. I did note two things:

1. There is a "desaturate" tool that offers the choice of preserving lightness or luminosity. Each gives very different results. Thus, the GIMP definition of these two terms could be different in GIMP.

2. To reduce color noise it seemed more effective to me to set the mode of the top layer to "value" and then blur the bottom layer.

I'm afraid I'm not very scientific when it comes to GIMP (or Photoshop)---I just keep hacking until it looks right...

gimp now uses 1 of 3 different methods for desat.
1) lightness
2) luminosity
3) average

pick one i normally use #2

The "desaturate" tool is all or nothing---in other dialogs having a controllable saturation, it seems to work on lightness (Which maybe explains the issue in the OP.)

In my experience, if your tests are giving you inconsistent results, you're probably working on an image that isn't very noisy. Try stretching the tonal range or pick a noisier file. The problem is very obvious when the noise is obvious.

I've tried all 3 modes of the Desaturate command and the results are as bad as in the tests I've posted. I create a new layer from the image, desaturate that layer, and set its transfer mode to Value. If everything worked as it should, you should see nothing, there should be no effect. Yet, the effect is very pronounced, with each of the 3 modes different, but none correct. The Desaturate command may be working correctly, but the Value transfer mode definitely isn't. It's difficult to make judgements here because the fact that the user has a choice of 3 algorithms actually serves to put the ball in the user's court: don't complain, you chose the algorithm. Well, one way or another, the right algorithm isn't available.

One good thing is that the new version coming up will apparently be using GEGL routines for more of its processing, so there's hope that might fix it.